Tricking numpy to see rows as values for a fake intersect2d

Numpy has a function called intersect1d which is quite helpful for finding intersections of lists of indices. So, for example, if you have two vectors of booleans you might want to find the intersection between them. This function doesn’t work when you have lists of n-d indices. This snippet shows how to trick numpy into seeing rows as values for a fake intersect2d.

Credit: User Joe Kington [StackOverflow]

import numpy as np
def intersect2d(a: np.ndarray, b: np.ndarray):
	if not (a.shape == b.shape):
		raise TypeError("a and b must have the same shape.")

	_, ncols = a.shape
	dtype = {
		'names': ['f{}'.format(i) for i in range(ncols)],
		'formats': ncols * [a.dtype]
		}
	return np.intersect1d(a.view(dtype), b.view(dtype))

A = np.array([[1,4],[2,5],[3,6]])
B = np.array([[1,4],[3,6],[7,8]])

print(intersect2d(A, B))
# [(1, 4) (3, 6)]